May 7, 2003 ©2001-2003 Howard Huang 1

PC I/O

Today wraps up the I/O material with a little bit about PC I/O systems.— Internal buses like PCI and ISA are critical.— External buses like USB and Firewire are becoming more important.

Today also happens to be the last day of class.

May 7, 2003 PC I/O 2

Open up and say ahhh

If you open up your computer, you’ll find the motherboard, which connects everything together.

May 7, 2003 PC I/O 3

The mothership

CPU socket

Memory slots

Serial,parallel,and USBports

(Back)

IDE driveconnectors

(Front)

AGP slot

PCI slots

May 7, 2003 PC I/O 4

What is all that stuff?

Different motherboards support different CPUs, types of memories, and expansion options.The picture is an Asus A7V.— The CPU socket supports AMD Duron and Athlon processors.— There are three DIMM slots for standard PC100 memory. Using 512MB

DIMMs, you can get up to 1.5GB of main memory.— The AGP slot is for video cards, which generate and send images from

the PC to a monitor.— IDE ports connect internal storage devices like hard drives, CD-ROMs,

and Zip drives.— PCI slots hold other internal devices such as network and sound cards

and modems.— Serial, parallel and USB ports are used to attach external devices such

as scanners and printers.

May 7, 2003 PC I/O 5

How is it all connected?

North Bridgechip

South Bridgechip

Modem Sound card

Hard disks CD-ROM

Videocard

Memory

CPU

AGPport

PCI bus

PCI slots

IDE controller Serial, paralleland USB ports

May 7, 2003 PC I/O 6

PCI

Peripheral Component Interconnect is a synchronous 32-bit bus running at 33MHz, although it can be extended to 64 bits and 66MHz.The maximum bandwidth is about 132 MB/s.

33 million transfers/second × 4 bytes/transfer = 132MB/s

Cards in the motherboard PCI slots plug directly into the PCI bus.Devices made for the older and slower ISA bus standard are connected via a “south bridge” controller chip, in a hierarchical manner.

North Bridgechip

South Bridgechip

33 MHz PCI bus

PCI slots

May 7, 2003 PC I/O 7

Fist of the North Bridge

The CPU and main memory exchange large amounts of data frequently.The 33MHz PCI bus is relatively slow for this, so a frontside bus is often dedicated to the CPU and main memory.— Some newer systems use 64-bit frontside buses running 100-166MHz.

By transferring data on both the positive and negative clock edges, the effective frequency increases to 200-333MHz.

— Other systems feature 16-bit buses running at 800MHz, and making four data transfers per clock cycle.

All of this goes through the “north bridge” controller, which can get very hot. The north bridge in the A7V is cooled by a green heatsink.

North Bridgechip

Memory

CPU

100-166MHz

64

64

May 7, 2003 PC I/O 8

Frequencies

CPUs actually operate at two frequencies.— The internal frequency is the clock rate inside the CPU, which is what

we’ve been talking about all along.— The external frequency is the speed of the processor bus, which limits

how fast the CPU can transfer data.The internal frequency is usually a multiple of the external bus speed.— A 2.66 GHz Pentium 4 might use a 533 MHz bus (533 × 5).— An 2.167 GHz Athlon XP sits on a 166 MHz bus (166 × 13).

Processors can often be overclocked to run faster than advertised.— Some motherboards allow you to change the external bus frequency,

the “clock multiplier” (e.g., 5 or 13), or both.— This can cause problems with some devices or the

CPU itself if you’re not careful or if you overclocktoo much.

May 7, 2003 PC I/O 9

Video cards are memory hogs

Graphics cards need a lot of fast memory for high-resolution images.— A 1024 × 768 resolution image in 32-bit color requires 3 MB of RAM.

1024 pixels/row × 768 rows × 4 bytes/pixel = 3 MB

— If the screen is refreshed at 75 Hz, this 3 MB image must be read 75 times per second.

Games also use textures, which are images that get pasted onto shapes. Textures can each require 256 KB of memory or more.

Chris Lattner,CS319 HW1

May 7, 2003 PC I/O 10

AGP

In the old days video cards used PCI, but that involved a slow 33MHz bus with a lot of contention and a maximum transfer rate of 132MB/s.New machines include an Accelerated Graphics Port.— The graphics card gets its own 32-bit, 66MHz path to the north bridge.— With clock doubling (AGP 2x), transfer rates can go up to 528MB/s

from the video card to main memory.The CPU and video card both need to access main memory quickly, so we still have possible contention problems.

North Bridgechip

Videocard

Memory

CPU

AGPport

528MB/s @ 66MHz × 2

32

800MB/s @ 100MHz

64

May 7, 2003 PC I/O 11

External buses

External buses can be much more convenient for end users.— PCI, ISA and AGP devices must be attached inside the computer. This

is difficult for some users, and laptops cannot even be opened.— External buses also allow devices to be connected from further away.

Some common external buses include serial and parallel connections, USB and Firewire.

May 7, 2003 PC I/O 12

Serial and parallel connections

A serial connection transmits data one bit at a time.A parallel port transmits multiple bits (typically 8) at a time. Although this can be faster, there are a couple of problems too.— Thicker cables are needed to

transmit more bits at once. Thiscan be a hindrance for smallerdevices like MP3 players.

— With many wires, interferencebecomes an issue. Special cablesmay be necessary for high-speedparallel connections.

— Delay or skew is also a problem,especially with longer externalconnections. All of the bits in aparallel transfer could arrive atslightly different times.

The serial USB connector is the small one,and the parallel connector is the big one.

May 7, 2003 PC I/O 13

USB

The Universal Serial Bus is a newer asynchronous serial bus standard, with many advantages over older serial and parallel buses.The USB 1.1 standard has a bandwidth of 1.5 MB/s, which is enough for keyboards, mice and small home networks.The newewr USB 2.0 standard boosts speeds up to 60 MB/s, which works better for faster devices like CD-ROMs or hard drives, which can transfer at rates of 7-40 MB/s.The bus can be shared by up to 127 devices. As usual, more devices leads to more bus contention, but most personal computers have only a couple of USB peripherals.

May 7, 2003 PC I/O 14

USB is friendly

USB has several other advantages.— It supports plug-and-play standards, so devices can be configured with

software instead of flipping switches or setting jumpers.— It also supports hot plugging, so you don’t have to turn off a machine

to add or remove a peripheral. This is important for portable devices that aren’t always connected to a PC, like an MP3 player or a PDA.

— The cable and connectors are small, with just four wires.— The cable transmits power! No more power cables, extension cords or

electrical fires needed!

May 7, 2003 PC I/O 15

Firewire

Firewire, developed and trademarked by Apple, is very similar to USB.It has three main advantages.— Most current implementations transfer up to 50MB/s, and the newest

versions of Firewire can go twice as fast.— No PC is needed! It’s possible to connect two devices directly, which

can make it easy to do things like copy data between two digital video camcorders.

— It provides more power to peripheral devices.These advantages make it attractive for video processing applications.

May 7, 2003 PC I/O 16

Common bus themes

Internal buses are taxed by fast devices that need large bandwidths, as well as many devices contending for the bus.— Clock doubling increases transfer rates with the same bus width.— Adding shorter and faster dedicated buses for high-bandwidth devices

like the CPU, main memory and video cards reduces contention.External buses have some different requirements.— Distances between devices are greater, so asynchronous serial data

transfers are often faster and more reliable.— Plug-and-play and hot plugging are important for end users.

Numbers are always just general guidelines! Most buses never reach their theoretical maximum bandwidth, for instance.

May 7, 2003 PC I/O 17

New bus trends

Emerging bus standards like 3GIO, HyperTransport and RapidIO combine many ideas from previous bus designs.— Serial connections avoid delay skews and interference, can run at high

speeds, and are less expensive to manufacture.— Using multiple buses or moving to “hub” or “switch” organizations can

reduce contention.— Data from several devices can be multiplexed onto a bus, much like

packets on a network. Thus, many devices can transmit data at once, instead of having to wait for previous transfers to complete.

New features like error detection, quality-of-service, and power saving are also important for applications like servers, real-time broadcasting, and embedded systems.

System bus

Hard disks CD-ROM Network DisplayCPU Memory

May 7, 2003 PC I/O 18

Instant replay

The semester was split into roughly three parts.— The first third covered instruction set architectures—the connection

between software and hardware.— In the middle of the course we discussed processor design. We focused

on pipelining, which is one of the most important ways of improving processor performance.

— Finally we talked about fast and large memory systems, I/O, and how to connect everything together.

We also introduced many performance metrics to estimate the actual benefits of all of these fancy designs.

MemoryProcessor

Input/Output

May 7, 2003 PC I/O 19

Some recurring themes

There were several recurring themes throughout the semester.— Instruction set and processor designs are intimately related.— Parallel processing can often make systems faster.— Amdahl’s Law quantifies performance limitations.— Hierarchical designs combine different parts of a system.— Hardware and software depend on each other.

May 7, 2003 PC I/O 20

Instruction sets and processor designs

The MIPS instruction set was designed for pipelining.— All instructions are the same length, to make instruction fetch and

jump and branch address calculations simpler.— Opcode and operand fields appear in the same place in each of the

three instruction formats, making instruction decoding easier.— Only relatively simple arithmetic and data transfer instructions are

supported.These decisions have multiple advantages.— They lead to shorter pipeline stages and higher clock rates.— They result in simpler hardware, leaving room for other performance

enhancements like forwarding, branch prediction and on-die caches.

May 7, 2003 PC I/O 21

Parallel processing

One way to improve performance is to do more processing at once.There were several examples of this in our CPU designs.— Multiple functional units can be included in a datapath to let single

instructions execute faster. For example, we can calculate a branch target while reading the register file.

— Pipelining allows us to overlap the executions of several instructions. — Superscalar CPUs permit parallel execution of entire instructions.

Memory and I/O systems also provide many good examples.— A wider bus can transfer more data per clock cycle.— Memory can be split into banks that are accessed simultaneously.

Similar ideas may be applied to hard disks, as with RAID systems.— A direct memory access controller performs I/O operations while the

CPU does compute-intensive tasks instead.

May 7, 2003 PC I/O 22

Amdahl’s Law

Performance is limited by the slowest component of the system.We’ve seen this in regard to cycle times in our CPU implementations.— Single-cycle clock times are limited by the slowest instruction.— Pipelined cycle times depend on the slowest individual stage.

Amdahl’s Law also holds true outside the processor itself.— Slow memory or bad cache designs can hamper overall performance.— I/O performance is also becoming increasingly important.

May 7, 2003 PC I/O 23

Hierarchical designs

Hierarchies separate fast and slow parts of a system, and minimize the interference between them.— Caches are fast memories which speed up access to frequently-used

data and reduce traffic to slower main memory.— Buses can also be split into several levels, allowing higher-bandwidth

devices like the CPU, memory and video card to communicate without affecting or being affected by slower peripherals.

May 7, 2003 PC I/O 24

Architecture and software

Computer architecture plays a vital role in many areas of software.Compilers are critical to achieving good performance.— They must take full advantage of a CPU’s instruction set.— Optimizations can reduce stalls and flushes, or arrange code and data

accesses for optimal use of system caches.Operating systems interact closely with hardware.— They should take advantage of CPU features like support for virtual

memory and I/O capabilities for device drivers.— The OS handles exceptions and interrupts together with the CPU.

Demanding applications rely on advanced system architectures.— The latest multimedia, database and supercomputing applications all

benefit from modern architectures.— Consumer devices such as cell phones and Gameboys frequently have

different requirements and need custom hardware and software.

May 7, 2003 PC I/O 25

Wait! Don’t stop now!

Those wacky electrical engineering people also have architecture classes.— ECE291 teaches Intel 8086 assembly language programming, and also

presents some architectural concepts along the way.— ECE312 is similar to CS232, but a little more advanced and with actual

processor simulations.The CS department has several higher-level architecture classes.

There are other related classes in our department also.

Computer System OrganizationCS333

Embedded Systems Architectures and SoftwareCS331

Compiler ConstructionCS326

Operating Systems DesignCS323

Programming Languages and CompilersCS321

May 7, 2003 PC I/O 26

Good luck on your exams and have a great summer!